Photo

Credit Victoria Roberts

Q. Can a computer tell from published documents whether the author is a man or a woman?

A. One computer program had a reported accuracy rate of 80 percent in its analysis of language patterns in modern written material.

In a study published in 2002 in the journal Literary and Linguistic Computing, researchers analyzed 566 published documents in British English using a program they called Winnow.

The program was “trained” on material that was labeled having been written by men or women. It then analyzed the comparative frequency in unlabeled documents of scores of features of writing judged to be independent of content.

The features included parts of speech, like nouns and pronouns, and function words, like “and” and “the,” which have little meaning of their own but indicate grammatical relationships within a sentence. It also studied patterns like two-and three-word phrases — for example, “above the table.”

The writings were fiction and nonfiction, and the same techniques can make that distinction accurately 98 percent of the time, the researchers said.

In the training phase, the authors identified as male features in fiction the more frequent use of “a,” “the” and “as.” Female features in fiction included “she,” “for,” “with” and “not.” In nonfiction, male features included “that” and “one,” and female features were “for,” “with,” “not” and “in.”

The underlying reasons for the differences and whether they would apply to other languages and cultures are subjects for further research. The program was far from flawless. One of the works it miscategorized as having been written by a man was the novel “Possession,” by A.S. Byatt, a woman. This email address is being protected from spambots. You need JavaScript enabled to view it.